On Tue, Jul 21, 2015 at 16:12:09 +0100, Daniel Berrange wrote:
On Mon, Jul 20, 2015 at 11:10:18AM +0200, Peter Krempa wrote:
On Fri, Jul 17, 2015 at 15:40:31 +0100, Daniel Berrange wrote:
On Thu, Jul 16, 2015 at 04:18:08PM +0200, Peter Krempa wrote:
...
The patch as-is removes the partition name from the code that generates the systemd SCOPE name since that does not contain it. The partition is used to generate SLICE name which is contained in the cgroup path before the SCOPE.
Ok, I understand what's going on now and reproduced the problem locally too. So ACK to your proposed patch.
I've pushed it now. Thanks.
The question that comes into my mind is whether we should actually make sure that the <partition> corresponds to the SLICE in the detected path, or just matching the machine SCOPE is good enough in this case.
I'm not sure it matters much - as long as we've detected the cgroup tree we'll do the right thing cleaning it up on guest shutdown, even if the <partition> in XML is different.
I have the same opinion. Additionally if we'd kill off the ability to configure cgroups just because that doesn't match it would be rather user-unfriendly. Peter